Kinetix

Kinetix (FLAIR, University of Oxford; ICLR 2025 oral) is an open-ended space of JAX-accelerated 2D physics RL environments. Procedurally representable levels cover robotic-style locomotion, grasping, and classic control inside one simulator, so a general agent can pretrain at scale and evaluate on held-out tasks.

Paper: Kinetix. Code: FLAIROx/Kinetix. License: upstream terms.

WorldFoundry is an official-results importer. Simulator execution, policy loading, and rollout collection stay upstream. The scorecard primary is success_rate.

Dimensions

• Open-ended 2D physics RL

  KindSuite

  JAX-accelerated procedurally representable levels: locomotion, grasping, classic control.

Data composition

• Size

  Oxford FLAIR open-ended JAX 2D physics RL environment space (ICLR 2025 oral). Simulator execution stays upstream.

• Sources

  github.com

• Splits

  realtime

• Modalities

  vision · action · state

• License

  upstream_terms

• Notes

  WorldFoundry imports official rollout exports. It does not run the JAX simulator or load policies.

Metrics

Metric

• success_rate

  DirectionHigher is better

  Primary episode success rate over evaluated rollouts.

  PrimarySuccess Rate

• task_success

  DirectionHigher is better

  Per-task success aggregate.

• episode_success

  DirectionHigher is better

  Per-episode success indicator aggregate.

• reward

  DirectionHigher is better

  Mean episode reward / return.

Leaderboard

Kinetix (arxiv 2410.23208) is an open-ended physics-control environment paper. The HTML has no numeric agent ranking table we can ingest; results are described qualitatively / in figures. Kinetix (2025).

How to run

worldfoundry-eval zoo benchmark-run \
  --benchmark-id kinetix \
  --mode official-validation \
  --official-results-path <official_results.json> \
  --output-dir tmp/kinetix/official-validation \
  --json

Default import: WORLDFOUNDRY_KINETIX_RESULTS_PATH. Optional root: WORLDFOUNDRY_KINETIX_ROOT.

In-tree pointer: worldfoundry/data/benchmarks/tasks/external/kinetix.yaml.

Outputs include scorecard.json (primary success_rate), raw_results.jsonl, and evaluation/summary.json.

← Benchmark Hub